Skip to contents

Changelog

irxforge (development version)

  • anonymize_dataset(): model_file is no longer passed to FeRx verbatim. The simulation runs on a temporary copy adapted to the anonymized dataset: the [data_selection] and deprecated [initial_values] blocks are removed, [data] path is rewritten to the absolute path of the simulation dataset, and [data] column maps are kept only when they rename a surviving covariate (maps onto or from structural columns, e.g. AMT = DOSE or DV = lndv, are dropped because the anonymized data already uses the canonical names). Simultaneous renames such as HTCM = WT with WT = HT are preserved, a relative [priors] from_fit path is made absolute against the original model’s directory, and # and // comments are handled as FeRx handles them (#23).

  • anonymize_dataset(): covariates are now checked up front for observed values, before any sampling or design matching. Every covariate needs at least one non-missing value overall and at least one on the baseline rows (each subject’s first record by time), which the samplers impute baselines from. All offending covariates are reported at once, and the error says “no observed values at baseline” when a covariate is observed only after baseline, instead of failing deep inside mice_impute() with “Variable HT has no observed values” (#25).

  • anonymize_dataset(): EVID == 0 records flagged MDV == 1 (e.g. scheduled samples that were never taken) no longer cause “FeRx simulation row count does not match the number of observation records”. Observations are now counted the way FeRx counts them (EVID == 0 & (is.na(MDV) | MDV == 0)); those records keep a missing DV and MDV = 1 in the output. If the counts still disagree, the error reports how many rows FeRx returned, how many EVID == 0 records there are, and how many of them are flagged MDV == 1 (#24).

  • sample_covariates_mice_timevarying(): fixed a crash (“Can’t combine ..1$.SEX_lag and ..2$.SEX_lag ”) when categorical covariates are stored as integer codes and at least one covariate is time-varying. Observed and simulated rows are now coerced to factors with a shared level set before stacking (#18).

  • anonymize_dataset(): the guard against time-reset records now tests what the time-varying samplers actually need — a non-decreasing TIME within every subject — instead of rejecting any dataset containing EVID 3 or 4. Reset records whose time scale keeps increasing are now accepted, and when the check fails the error names the offending subjects (#21).

  • anonymize_dataset(): covariates may now be character(0) for datasets without covariate columns. Covariate sampling is skipped; subjects are still renumbered 1:n (with each observed design cloned onto a new subject in random order) and DV is still replaced by simulated concentrations (#20).

  • check_nm_dataset(): fixed a crash when TIME is non-numeric (e.g. date/clock strings) and ADDL/II are present — the overlapping-dose check now also guards on is.numeric(TIME) so the whole check no longer aborts before returning results.

  • check_nm_dataset(): the unique_records check now skips (returns NA) instead of emitting a false FAIL when the subject ID does not resolve or TIME is not numeric (a character clock-of-day TIME, or a key collapsed across subjects, produced spurious duplicates).

  • check_nm_dataset(): failing char_columns, reserved_names, and covariate_missing checks now name the offending column(s) in their description.

  • Initial CRAN submission.